MQA is Vaporware

[Jim Austin]

13 hours ago, crenca said:

"Post-Truth" is very apt.  I would invite @John_Atkinson, @ARQuint, or perhaps @Jim Austin to explain, even just a little bit, how Bob S has discovered something fundamentally new in signal processing - something that is truly "Post-Shannon".

 

In fact, at the risk of appearing "extreme", I dare you - no I DOUBLE DARE you to give any credence to this claim at all.  

 

 

Hey, thanks for the invite. 

 

I encourage you to read the article--my little interview. I mean actually read it--because it says most of what I have to say, and it doesn't say what you seem to think it says. This is a problem you guys have, I notice: You question everyone's motives (and competence) and assume everything is freighted with conspiracies and hidden meanings. 

 

I'm not shocked to discover that you seem to think sampling theory ended with Shannon, because you smugly assume you know everything. It didn't. Shannon's work was remarkable and brilliant and extremely important--decisive for digital audio--but it was almost an afterthought for Shannon, a thing that needed doing so that he could accomplish some other thing (I forget exactly what). Even when Shannon's paper was published, others were already moving the field forward. Post-Shannon sampling theory is a real thing. 

 

Does it circumvent Shannon? No. Shannon was correct, so his theory can't be "fixed." Did I claim it did? Did I imply it? Since you folks seem to be hard of reading, or of honest thinking, I'll answer my own question: No, I did not. Did Bob Stuart make that claim? Not that I'm aware of, at least not in our interviews. You still cannot perfectly reconstruct a non-band-limited signal.

 

But the post-Shannon formulation does reframe things in interesting and important ways--and in science (including applied science), reframing issues often leads to new advances. Plus, once you realize there's a post-Shannon formulation--that it exists--it's sensible to stop worshipping at the Nyquist-Shannon altar and rethink things. You can't understand what's happening here without learning something about post-Shannon sampling theory. So, go read some Michael Unser; he has (at least) two papers on the subject. You'll need to know some math. 

 

Post-Shannon intrinsically incorporates the fact that most signals you want to convert are not strictly band-limited--hence, do not satisfy Shannon's most fundamental criterion. To make it satisfy that criterion, you have to do destructive filtering--so why not factor that into the theory directly and count your errors on both sides of the ledger? You replace the myth of perfection with an approach that balances time-domain errors with frequency-domain errors--a more honest accounting. So you think it through, you figure out the implications, and you conceive a new approach to audio just as others have conceived a new approach to several areas of image-related science. Why not? 

 

When you generalize Shannon, you get end up with a new mathematical formulation with deeper symmetries. It leads to a wider range of possibilities for sampling and reconstruction functions, including possible replacements for  the sinc function, which you folks view as the only correct way of doing things but the imaging people have recognized for decades as defective because it's very long in the time domain--sinc(x) extends out to infinity and damps out slowly. Fortunately it can be replaced by new functions including combinations of splines--which are, if memory serves (I read this months ago) the answer to the question, "what's the shortest set of functions subject to the appropriate constraints?" You really should do your homework before trying to destroy peoples' careers. 

 

Rethinking Shannon has made a huge difference in several areas of imaging science. Real, practical advantages have resulted. Take a look at the work of Michael Unser. Could an application of post-Shannon sampling theory do the same in audio? It appears that no one has tried it before now. So why not try it? 

 

The main problem with MQA in the eyes of folks like you is that it's proprietary and not 'open'. That's fine--a reasonable objection, as I wrote last month. But beyond that you're being dishonest--perhaps even with yourselves.  When someone points out that Shannon's may not be the best standards to judge by, you react indignantly like someone has just questioned Newton's Second Law, or gravity, and keep pointing out that it violates Shannon.

 

Sure, there is a chance that Bob Stuart is making this up. Post-Shannon sampling theory is real, an potentially important, but maybe they're not actually using it. Maybe he's lying in that interview (although the formulation outlined in publications and patent applications fits). I don't have the access (nor, perhaps, the expertise) to ensure that he's implementing a new codec based on a rethinking of Shannon. But to judge whether he's doing that or not, you need need to consider that it might be real. You haven't done that. When I consider who's likely to be more reliable--AES Fellows with long and distinguished careers or some self-important, anonymous Internet trolls with a cheap audio interface and Adobe Audition--the answer seems obvious to me. 

 

The question you ought to be asking is how much difference this makes in an audio context. I think it's a better way of doing things--but is it enough better to justify the overhead? That's the crux of the issue--not whether it's all fake.

 

I've given you what you asked for, now I'm done. Arguing with you folks is pointless. I've already wasted an hour or so of my life writing this; I'm not going to do it again. I prefer a more open-mined audience. Besides, it seems that anyone on this forum who dares go against the prevailing view ends up getting banned. 

 

 

I'm out. 

 

 

jca

[Jim Austin]

5 minutes ago, psjug said:

So technical arguments against MQA are off the table?

 

BTW I do not share the "good riddance" sentiment.  I think it would be good to engage in cool-headed discussion if you are willing.

 

I'll answer you because you're considerate. No, technical arguments against MQA are not off the table. But those who criticize it on technical grounds should endeavor first to understand MQA, and like any good scientist, should carefully consider opposing points of view instead of using cheap rhetorical techniques to de-legitimize it. 

 

To address a couple of other points: 1. I'm not an "academic". 2. That we don't need MQA is a valid opinion, but only that. The other answers to my post--the ones I noticed--are substance-less and inflammatory--an possibly in violation of the CA rules. Not that I particularly care. Do your worst. 

 

jca

[Jim Austin]

I used to be an academic, years ago. I'm not anymore. Why is this hard to understand? 

[Jim Austin]

9 minutes ago, Brinkman Ship said:

Your background is impressive and accomplished ...

 

Thanks. 

 

10 minutes ago, Brinkman Ship said:

which makes it impossible to understand how you have fallen under the spell of Charlatans like Bob Stuart.

 

It's at times like this that you need to reconsider your previously unquestioned assumptions: "Gosh, maybe I was wrong. Maybe there's something to this after all." I've seen too many of your posts to expect it though. 

 

12 minutes ago, Brinkman Ship said:

In the end the only answer one can find is commerce.

 

Not true. First, it's wrong to consider me an MQA advocate. I am, rather, someone whose background gives them (me) a certain obstinacy when it comes to accepted norms, a tendency to reject bandwagons, to question what everybody else thinks they know. That obstinacy is paired with a deep commitment to logic and actual evidence. I think, does it really make sense that two scholars would risk their reputations by becoming charlatans late in their lives? I don't think so--so then it makes sense to dig in and try to figure out what they're really up to. It's hard, because it's proprietary and they haven't been forthcoming--but that doesn't mean their tech is bullshit, as you and some others maintain. This is the foundation of what I write: a desire to air out the details, to push back against the zealots who are either ignorant or blinded by their zeal. The result, I hope, is an even-handed look at the technology, its problems, and its promise. 

 

Your inability to entertain the possibility that there's something to MQA makes it obvious that you are not thinking clearly. The facts call for reevaluation. Are you up to it? 

[Jim Austin]

5 minutes ago, Indydan said:

 

Have you reevaluated your position with some of the facts presented here (Archimago's technical paper)? 

 

I haven't looked at it lately, but my recollection is that there was nothing in it that was not already known.

[Jim Austin]

1 minute ago, mcgillroy said:

Jim I appreciate you posting. Three questions: have you seen the math, done the math, can you show the math?

 

You're welcome.

 

If you mean the math behind post-Shannon sampling, the answers are Yes, no, and no. I've read two papers as best I can, but it has been a long time since I did serious math--they are beyond me. If you mean, have Craven and Stuart shown me their work, I answer 'no' to all three. I've interviewed Stuart extensively--three intervals in total I think, and many emails, and I've had the luxury of the occasional MQA-encoded test track. But the only technical material I've seen is what's widely available in patent applications and scientific journals. But it holds together, and as I've said, Stuart and Craven are not charlatans. They're serious people with serious careers behind them--and I'm not talking about Meridian. In contrast to many hear, Stuart stands up in front of packed rooms and stands behind his claims. That doesn't mean he isn't lying, but to me the preponderance of evidence suggests otherwise. 

[Jim Austin]

2 minutes ago, mansr said:

If you already knew all that, why have you been ignoring it?

 

I don't know what you mean. What's to ignore? What's so damning? That leaky filters are leaky? That minimum-phase "causal" filters have non-linear phase response? 

[Jim Austin]

16 minutes ago, mcgillroy said:

Btw: I wrote Unser, let’s see what he has to say.

 

Forgot to mention: I too wrote to Unser. He didn't answer me. 

[Jim Austin]

1 minute ago, adamdea said:

So, then. Cutting to the point- where is the evidence of

a) an audio signal

b) an audible audio signal

which required post-shannon sampling to encode and reproduce? If you can answer that you have a moved one part of the debate along. If you can't  then you have either not asked the right questions or you don't care about the answers.

 

I've made these points more than once in my Stereophile writings. Two things need to be clearly established: That MQA improves impulse response in the real world--which may manifest itself in the signal envelope, not necessarily in very fast signals--and that the difference is audible and preferred. Because of its overhead, if it can't meet those two criteria, it's probably not worth supporting. We have not yet managed to do the first test (which requires cooperation from MQA Ltd), and preliminary word from the McGill test is NOT reassuring. The article they wrote will be presented later this month at an AES meeting in Milan. 

 

I said I wasn't going to post more, but I did. There are two reasons not to: I've got work to do--this is a time-sink--and the goal of some on this forum is to suck me in then rip my words apart. Importantly, there's no interest in learning anything, so there's no point in my continued participation. But to those of you who have been courteous, thanks. 

 

jca

[Jim Austin]

What are you folks still on about? Still assuming that all sampling theory ended with Shannon? This place is obviously a massive circle-jerk, but your posts are visible to the outside world, so you run the risk of embarrassing yourselves (those of you not hiding behind a pseudonym, that is). So I'll do you a favor--I'll present some quotes from scientific literature. Then you can dig them out yourself if you care to educate yourself; I don't expect that, but maybe it'll at least convince you to think twice before posting ignorant things. 

 

From an article by Yonina C. Eldar and Tomer Michaeli, published in IEEE Signal Processing Magazine. Both are at Technion, an the senior author--Eldar--has a PhD from MIT.
 

Quote

... Many

of the limitations encountered in current converters is due

to a traditional assumption that the sampling stage needs to

acquire the data at the Shannon-Nyquist rate, corresponding

to twice the signal bandwidth [6]–[8]. To avoid aliasing, a

sharp low-pass filter (LPF) must be implemented prior to

sampling. The reconstructed signal is also a bandlimited

function, generated by integer shifts of the sinc interpolation

kernel.

 

A major drawback of this paradigm is that many natural

signals are better represented in alternative bases other

than the Fourier basis [9]–[11], or possess further structure

in the Fourier domain. In addition, ideal pointwise sampling,

as assumed by the Shannon theorem, cannot be

implemented. More practical ADCs introduce a distortion

that should be accounted for in the reconstruction process.

Finally, implementing the infinite sinc interpolating kernel

is difficult, since it has slow decay. In practice, much simpler

kernels are used, such as linear interpolation. Therefore,

there is a need to develop a general sampling theory that

will accommodate an extended class of signals beyond bandlimited

functions and will account for the nonideal nature

of the sampling and reconstruction processes.

Sampling theory has benefited from a surge of research

in recent years, due in part to the intense research in wavelet

theory and the connections made between the two fields.

In this survey, we present several extensions of the Shannon

theorem that have been developed primarily in the past two

decades, which treat a wide class of input signals as well

as nonideal sampling and nonlinear distortions. ...

 

If you're feeling ambitious, look up "Sampling Moments and Reconstructing Signals of Finite Rate of Innovation: Shannon Meets Strang–Fix," by Pier Luigi Dragotti et al.; you can find the full citation in the MQA literature. The gist is that under appropriate conditions, signals that have a finite number of degrees of freedom per unit of time, "such as, for example, nonuniform splines or piecewise polynomials," can be perfectly reconstructed even if they are not band-limited ... but you must use appropriate sampling kernels--not the usual sincx function. The class of kernels "that we can use is very rich and includes functions satisfying Strang–Fix conditions, exponential splines and functions with rational Fourier transform." I'm not sure this strictly applies to a music signal, but in practical terms that may not matter--especially since there's no real need to recreate the waveform perfectly. 

 

Finally there's this, by Michael Unser, from IEEE Signal Processing Magazine. My copy of this is in pixels, not characters, so I'll paste in the image: 

...

There's more I could quote. 

 

None of this proves that MQA is a good idea, that it's valid mathematically, or that its application to music makes sense, and it certainly doesn't prove that MQA sounds better. It does however show what some of you still seem to be questioning: sampling theory didn't end with Shannon. The body of theoretical work referred to in the latest MQA article/interview is real. 

 

So stop making fools of yourselves by pretending it doesn't exist. Or, just keep making fools of yourselves. Makes no difference to me. 

 

Oh, and no, I won't provide the citations. If you want to read them you can find them on your own. 

 

jca

[Jim Austin]

Just now, The Computer Audiophile said:

 

Keep digging Jim. The hole is only getting deeper.

 

 

So, quite obviously, you're now abandoning any pretense of objectivity and even-handedness. 

[Jim Austin]

On 5/14/2018 at 2:17 PM, mansr said:

I must have spent hundreds of hours reverse engineering MQA for some other reason then.

 

I thought I'd go ahead an answer a few posts. I can't keep doing this however; there's simply not enough time, and only a few things here are worth responding to. 

 

It's apparent to me that you spent hundreds of hours reverse-engineering MQA in order to try to prove that it's invalid. 

[Jim Austin]

On 5/16/2018 at 1:26 AM, Fokus said:

First, I assure you that BS is quite capable of dreaming this up.

 

Second, he was not alone in doing this: there has been a decades-long cooperation with Peter Craven (who brings a lineage going back to Michael Gerzon). Ages ago Craven and Stuart started a war with orthodox steep linear phase reconstruction filters. This informed the design of Meridian CD players and DACs for a while. MQA is just the next step, getting rid of the filters altogether.

 

Exactly right. And for what it's worth, Charles Hansen, who is now a hero to any here, embraced similar design philosophies. (Hansen was a brilliant designer He deserves praise. I'm just pointing out the hypocrisy here.) 

 

A more important point is that people here point to "experts" who agree with them--and many of them are very knowledgeable. I have great respect for (to name two with whom I've privately discussed MQA) John Siau and Bruno Putzeys). Both are anti-MQA, and very knowledgeable. It's clear from these and other conversations, though, that the ideas behind MQA are not in the normal digital-engineer curriculum. You've got to go beyond that to evaluate them properly. My argument has consistently been that there's more to them than critics (like those here) give credit for. They can't be dismissed as facilely as many try to do. Do the homework first, then dismiss if it doesn't hang together. But keep an open mind. 

 

Goes without saying that you cannot judge it if you don't understand it. 

 

jca

[Jim Austin]

On 5/15/2018 at 1:57 AM, Fokus said:

That it replaces free and open PCM with closed-system packaged PCM, and adds the need for special hardware to boot. Without tangible benefit for the consumer.

 

 

Yes to your first point--and that is (as I've written) a legitimate concern. (It is not, as I have also written, something I personally worry about much.) As to the second, we have to wait and see. That's what listening tests are for. Given the stakes, this is not something that should be left to some self-proclaimed golden-eared writer--or me. It should be tested properly--and how it sounds is the ultimate test. And as I mentioned previously, word is the McGill tests didn't go well for MQA. If that proves to be true--if MQA (at comparable rate) is found not to be preferred over 24/96 PCM (the comparison they're making)--then the only advantage is some streaming economics. That, to me, is far less compelling. 

[Jim Austin]

On 5/17/2018 at 5:31 AM, The Computer Audiophile said:

2) Old guard print articles calling MQA the best thing to ever happen in digital and repeating falsehoods?

 

It took me longer than it should have to realize what you're on about. You see yourself in competition with magazines like Stereophile and TAS. So there's a bias against ideas espouse by those magazines. 

 

I am curious which "falsehoods" you think are being repeated. Can you cite them? Can you definitively show them to be false? 

 

It appears that your fiduciary conflict of interest compromises your objectivity. 

[Jim Austin]

On 5/17/2018 at 10:36 AM, The Computer Audiophile said:

I recently heard something so surprising that I'm trying to back it up with other sources. Trust me, if I can back this up, it isn't good news for MQA.

 

If it's the news about the McGill test, there's an abstract already on the AES website. I've only read the abstract--haven't seen the paper--but I've talked to an expert who has read the paper. If this is your news, I'll break it: The listening tests at McGill failed to establish a difference between MQA and PCM at the same rate. That's an impressive showing for MQA's compression, but IMO it devastates their value proposition. 

 

In other news, the leading producer of DAC chips will incorporate MQA natively into their DAC chips. That's big news, too, but that's been public for a while, so that's probably not it. 

[Jim Austin]

On 5/17/2018 at 7:52 PM, crenca said:

In any case, "post shannon" is a spin term meant to describe a certain philosophy about filtering, audibility/desirability of "ringing", IM, and the like that is all based on shannon.  It's more of the same from the MQA promotion machine...

 

Then why was it widely use in the late 1990s? 

[Jim Austin]

On 5/18/2018 at 5:34 AM, adamdea said:

The odd thing about Jim Austin's sentence is that the frequency domain and the time domain are mathematically interchangeable. The Shannon's proof of the sampling theorem depends on this. That relationship is immutable. So how can they not be symmetrical and how can anything restore what can;t be lost?

 

Because music in the analog domain is not inherently band-limited. So, one has to apply an antialiasing filter pre-conversion, thus altering the signal that will be "perfectly" reconstructed (ignoring some complications related to amplitude quantization). 

 

It is in the assumption of bandwidth limitation that there is an implicit lack of symmetry. One counts errors in the frequency domain while ignoring errors introduced by antialiasing. In the generalized post-Shannon approach, those errors are counted--integrated into the theory. 

[Jim Austin]

On 5/18/2018 at 6:01 PM, mansr said:

The Shannon-Nyquist sampling theorem provides a sufficient condition for fixed-interval sampling to fully capture a signal and enable subsequent reconstruction. Later research has defined other conditions allowing certain signals to be accurately captured without fulfilling the Shannon-Nyquist criterion. A search for terms like sparse signal, sparse sampling, compressed sensing, and finite rate of innovation will turn up hundreds of papers spanning decades.

 

None of this is new. The reason it hasn't been applied to audio is that it is unnecessary. An audio signal has such a low bandwidth to begin with that the Shannon-Nyquist requirement is easily met. The data rates involved also pose no problems for processing, transmission, or storage systems. Even if some form of sparse sampling of audio could cut the data rate in half, say, there are good reasons not to do this outside very specific applications. Traditional sampling produces a signal that is easy to process in a multitude of ways (think of all the operations a DAW can do). That isn't necessarily true of sparse sampling. Why should we complicate everything only for the sake of a data rate reduction we don't need? I can't think of a single reason.

 

When, months ago, I started exploring this topic, I did the same Google searches you've apparently done. Yes, the most obvious application of non-Shannon sampling is when you've got a signal with high and low limits. But when you keep reading, you realize there's much more to it than that--but that it has not been applied much to signal processing. 

[Jim Austin]

19 hours ago, The Computer Audiophile said:

 

 

 

Exactly right. 

[Jim Austin]

3 hours ago, Ralf11 said:

The real question here is whether Jim Austin misunderstood the papers he refused to cite, or if he is actively engaged in a witting deception.

 

Do you really feel that you understand the topic well enough to make that accusation--that I'm either ignorant or lying? 

[Jim Austin]

3 hours ago, miguelito said:

This is key. I went to a demo in March 2015 and the MQA versions were quite better. But looking back and after all of the experience listening and comparing albums on TIDAL (MQA, TIDAL redbook, ripped redbook, and high res versions I own), it is pretty clear that the MQA gains in some albums are entirely due to remastering and not to MQA itself, and as such it could clearly be accomplished with PCM alone.

 

There certainly are plenty of MQA tracks/albums floating around whose CD-res versions are from different masters, but I have not found that to be the case for high-res versions; generally they appear to be from the same masters. To charge that using different masters in their demos is to suggest that they are engage in the worst sort of fraud. Even if you think them ethically challenged, it's illegal and extremely dangerous strategically. For that reason, I think it unlikely. 

[Jim Austin]

36 minutes ago, beetlemania said:

True that Ayre’s listen filter is similar to MQA’s filter. But I don’t recall that Ayre intended to engineer a closed system that all would have to adopt, from recording to user. Only one of your articles has addressed some of the downsides of MQA, and gently at that. I kindly suggest you interview some of the MQA critics for future articles. 

 

I have talked to several. Some do not want to go on the record. Others have signed noncompete agreements. Have you noticed that even on this forum most experts use pseudonyms? Doesn't make it impossible however, an it's a reasonable suggestion. 

38 minutes ago, beetlemania said:

Meanwhile, many of us have compared MQA to true hi-res and we’re not impressed. 

 

Fine--good. Thanks for listening with your ears. 

[Jim Austin]

4 hours ago, miguelito said:

Just consider reading Bob's incomprehensible gibberish vs @mansr clear statements. Really, it is not that Bob is speaking at a more complex level. It is gibberish.

 

Do you think you can make a convincing case that this is true? It's a pretty serious accusation. Since some of this work has been published in scientific journals--I mean Bob Stuart's--you're accusing him of scientific misconduct, among other things. I'll point out again that his work has earned him distinction as a Fellow of the Audio Engineering Society. You are welcome to your opinion, but even without such distinction, there should be a high bar for that, IMO. 

[Jim Austin]

33 minutes ago, Mordikai said:

So you understand this stuff but Bruno Putzeys doesn't? Yawn, this is basically trolling at this point.

 

Hardly. Bruno has forgotten more about digital audio than I'll ever know, and as I've said before, I don't understand it either. The point is that as I've learned, it's not necessary for a digital engineer to understand this stuff. If it is what it claims to be, MQA is based on a rethinking of what digital audio engineers consider gospel. The question is whether it makes the music sound better.